March 17, 2026 at 2 PM
Soft robots offer unique advantages for contact-rich interaction due to their compliance and morphological adaptability. However, their high-dimensional deformation, multi-physics coupling, and contact-dependent inverse dynamics make real-time control and scalable learning significantly more challenging than in rigid systems. As a result, much of soft robotics research has focused on modeling and low-level control, while generalizable interaction learning remains underexplored. In this talk, I will present a structured framework that bridges real-time multi-physics modeling, inverse control, and learning-based interaction. First, I introduce a unified and stabilized dynamics formulation that enables real-time forward and inverse computation for continuum soft robots. Second, I describe a learning-oriented simulation infrastructure that supports standardized trajectory and contact-state logging, large-scale rollouts, and human-in-the-loop data collection. Finally, I discuss how low-level interaction skills and structured inverse models can interface with Vision-Language-Action (VLA) policies, enabling task-level generalization while maintaining stable execution in contact-rich settings. Overall, this work explores how physics-grounded infrastructure can support scalable and reproducible soft-robot interaction research.
français
English